General comments

This paper describes a new measurement technique for identifying pollen grains that combines holography and fluorescence.  Overall, the paper reports methods and laboratory results that will be useful to the pollen observation community, although with the caveat that this has only been tested and trained in the lab and has not yet been used on ambient samples.  It would be helpful to include something to this effect (e.g, laboratory only, needs to be tested in ambient air) in the abstract, as it only is raised briefly in the discussion.  I would have preferred to see some ambient samples presented in the paper but recognize that this increases the scope of work substantially.  Otherwise, the technique holds promise for distinguishing different types of pollen for real-time sampling, which is an exciting result.  I have several minor presentation comments below that would make this manuscript acceptable for publication. 

Minor comments

1. Title: “measurement” -> “measurements”
2. Line 61: “… for the main allergenic species….” – is this the main species in Switzerland, or more broadly in Europe?  More clarity for the selection of these seven pollen types would be beneficial.
3. Line 75: What size of particles trigger the detector? While this study is specifying the types of pollen evaluated, what would happen if it were an ambient air sample?
4. There is some terminology in the paper that is rather confusing:
   1. “event” – this is defined on line 89, but that makes it sound like you are sampling ambient air versus a controlled emission. Also confusing in Table 1, where it really seems to be the number of particles counted and evaluated. I would suggest something like “number of particles counted”, “number of images”, or even just “pollen count”
   2. “class” – defined on line 95 as the same as plant taxa, although it was unclear why this specific term was used – why not just keep “taxa”? Also, later throughout the paper (e.g., line 218, 241, y-axis label on Figure 5) these are used interchangeably and makes it rather confusing. 
5. Table 1: If using the term “class” in the paper, I would suggest changing the header on the first column from “Common name” to “Class (common name)” to clarify the terms more clearly.
6. Line 179: eccentricity seems to be a useful metric, but is there also one about symmetry that might be helpful for more complex grains?
7. In Figure 3c, the standard deviation on the relative fluorescence is extremely large. Can the authors comment on why they think that this metric improves the ML model?
8. Line 183: “superior” -> “larger”
9. Line 247-248: Please rephrase

This manuscript evaluates the effectiveness of including fluorescence information along with holographic images for the identification of a small number of allergenic pollen species. They find that the accuracy of the combined technique is a large improvement over the use of only fluorescence or only holography for identification. While this is a nice piece of work that shows value in this combination, I do think it is quite limited in applicability due to the small number of pollen samples examined. A few questions/issues that I hope the authors can address are as follows:

I would recommend a bit more instrument description early on in the paper. I realize that the instrument has been described previously but it would be helpful to have at least a brief summary of the most pertinent details presented here since it is not an instrument that has been written about extensively and may be unfamiliar to some readers. It would be nice to know the detectable size range, what the sizing laser and the fluorescence excitation sources are, how the excitation is triggered by the instrument and how the scattered/emitted light is collected.

Along similar lines, I think it would be good to describe the data pre-processing in section 2.2 a little more clearly. I believe you mean that a human user removed both non-pollen particles and any pollen particles that were not of the particular class being sampled. If I’m interpreting that correctly, that seems like quite a limitation. If the plan is to eventually use this technique on ambient data the instrument and the ML methodology is going to need to be able to deal with both non-pollen particles and sampling periods in which many different kinds of pollen are mixed together. If this cleaning were not done, how much worse would the performance be? Can you comment on whether this would be a necessary step in processing ambient data and, if so, how that might be accomplished and what it would mean for identification of different pollen types?

The authors state that, if subtraction of the baseline resulted in apparently negative fluorescence values, those values were assumed to be zero instead. Although this is likely a minor effect, I think it is more mathematically correct to include negative values as, without them, you are necessarily biasing the data high. Aren’t they statistically meaningful in the sense that they reflect baseline variability in the instrument?

The authors are careful to sample fresh pollen quickly which is appropriate for a laboratory study. I believe, however, that there is reasonable evidence from recent measurements (see Hughes et al., ES&T Lett. 2020 for example) that pollen ruptures into smaller particles when exposed to humid conditions. It seems that the size and shape parameters are quite important for telling these 7 pollen types apart, could the authors comment on how this method might perform if there were pollen fragments in atmospheric measurements?

Generally I think the limitations of the controlled nature of the study and the small number of homogeneous samples used, should be discussed more clearly throughout. Although this offers a nice proof of concept that fluorescence may help discern different pollen types from one another, there is still the potential that a more diverse and variable ambient particles population will not be so readily separated. It would have been interesting, rather than focusing on the most allergenic species to, instead, focus on the most common species as a function of season as that would be more likely to tell you whether this method is capable of giving actionable information to allergy-prone citizens.

Line 217 – I think you mean “averages” rather than accuracies

For Figure 4, is that based on one of the two instruments used in the study? If so which one? How different would the discernment be if the other instrument was used?